scheduler

Quicker schedule generation for classtime.

Feature roadmap:

• represent Section as a concrete type (see rosshamish/classtime#103)
• represent Schedule as a concrete type
• score schedules
  • with len(Sections)
  • with rough ports of the Python scoring functions
  • with scoring functions whose scores are bell-curved (ish)
• get section data from course ids
  • directly from psql
  • from temporary classtime api endpoint, /api/vx/sections
  • ?
• conflict detection
  • same course, same component
  • same time
  • course dependency not satisfied (ie LECA1->LABA2,3,4)
    • this is half-implemented, need to check the AutoEnroll component type, see the Python impl
• gen with constraint solver (SAT solver)
• busy-times
• more progressive conflict detection (see rosshamish/classtime#109) SAT solver makes these unnecessary
  • is this pair of sections a known conflict? -conflict
  • are these sections on different days? -noconflict
  • do their times overlap? -conflict
• conflict data collection
  • known conflicting section pairs per course pair
    • "do these courses fit pretty well together? Or will this schedule be a lost cause with this pair?"
    • this could lead to course suggestions based on what you've already got in your schedule
  • return schedules with 1 or 2 conflicts but are great schedules
• caching
  • of known conflicting section pairs
  • of schedule requests
  • persistent across instances & restarts
• electives
• "more like this" condensing
• expose as an API with same interface as classtime/api/vN/generate-schedules
• redirect classtime/api/vN+1/generate-schedules requests to this API

Benchmark roadmap:

comparing against ~30s for sequentially running all cases at TestAPI.test_generate_schedules

• running one test properly with bare minimum
• running properly with busy-times
• running properly with electives too it's broken in classtime right now anyway
• running all tests properly
• 90s
• 45s
• 30s
• 15s
• 10s
• 5s
• 1s